Rockhead.Extensions 0.1.0-preview

Rockhead.Extensions

Rockhead.Extensions is a set of extension methods for the AWS SDK for .NET Bedrock Runtime client. Amazon Bedrock is a serverless service making available through a unique API foundation models (FMs) from Amazon, AI21 Labs, Anthropic, Cohere, Meta, Mistral AI and Stability AI.

Rockhead.Extensions provides you extension methods to the low-level Bedrock Runtime API with strongly typed parameters and responses for each supported foundation models. It makes your developer life easier.

Rockhead.Extensions

Available on Nuget: https://www.nuget.org/packages/Rockhead.Extensions

• AI21 Labs extension methods
  • InvokeJurassic2Async
    • An extension method to invoke Jurassic 2 models to generate text with strongly type parameters and response
    • Support Jurassic 2 Mid or Ultra models
• Amazon extension methods
  • InvokeTitanImageGeneratorG1ForTextToImageAsync
    • An extension method to invoke Titan Image Generator G1 to generate images with strongly type parameters and response
  • InvokeTitanTextG1Async
    • An extension method to invoke Titan Text G1 models to generate text with strongly type parameters and response
    • Support Titan G1 Lite or Express
  • InvokeTitanTextG1WithResponseStreamAsync
    • An extension method to invoke Titan Text G1 models to generate text with strongly type parameters and returning an IAsyncEnumerable of strongly typed response
    • Support Titan G1 Lite or Express
  • InvokeTitanEmbeddingsG1TextAsync
    • An extension method to invoke Titan Embeddings G1 models to generate embeddings with strongly type parameters and response
  • InvokeTitanMultimodalEmbeddingsG1Async
    • An extension method to invoke Titan Multimodal Embeddings G1 models to generate embeddings with strongly type parameters and response
• Anthropic extension methods
  • InvokeClaudeAsync
    • An extension method to invoke Claude models to generate text with strongly type parameters and response
    • Support Claude Instant v1, Claude v2 and Claude v2.1
  • InvokeClaudeWithResponseStreamAsync
    • An extension method to invoke Claude models to generate text with strongly type parameters and returning an IAsyncEnumerable of strongly typed response
    • Support Claude Instant v1, Claude v2 and Claude v2.1
• Cohere extension methods
  • InvokeCommandV14Async
    • An extension method to invoke Command v14 models to generate text with strongly type parameters and response
    • Support Command v14 Text and Command v14 Light Text
  • InvokeCommandV14WithResponseStreamAsync
    • An extension method to invoke Command v14 models to generate text with strongly type parameters and returning an IAsyncEnumerable of strongly typed response
    • Support Command v14 Text and Command v14 Light Text
  • InvokeEmbedV3Async
• Meta extension methods
  • InvokeLlama2Async
    • An extension method to invoke Llama 2 models to generate text with strongly type parameters and response
    • Support Llama 2 13B Chat v1 and Llama 2 70B Chat v1
  • InvokeLlama2WithResponseStreamAsync
    • An extension method to invoke Llama 2 models to generate text with strongly type parameters and returning an IAsyncEnumerable of strongly typed response
    • Support Llama 2 13B Chat v1 and Llama 2 70B Chat v1
• Stability AI extension methods
  • InvokeStableDiffusionXlForTextToImageAsync
    • An extension method to invoke Stable Diffusion XL to generate images with strongly type parameters and response

Setup

• Rockhead.Extensions package is available on NuGet: https://www.nuget.org/packages/Rockhead.Extensions
• For now, Rockhead.Extensions is only available for .NET 8
• You can install Rockhead.Extensions with the dotnet CLI:

dotnet add package rockhead.extensions

Usage

Below are a few examples of using the extension methods to invoke different models.

Invoke Claude v2.1 model

public async Task<string> GetLlmDescription()
{
    const string prompt = @"Human: Describe in one sentence what it a large language model\n\nAssistant:";
    var config = new ClaudeTextGenerationConfig()
    {
        MaxTokensToSample = 2048,
        Temperature = 0.8f
    };
    
    var response = await BedrockRuntime.InvokeClaudeAsync(new Model.ClaudeV2_1(), prompt);
    
    return response?.Completion ?? "";
}

Invoke Claude v2.1 model with a response stream

public async Task<string> GetLlmDescription()
    {
        const string prompt = @"Human: Describe in one sentence what it a large language model\n\nAssistant:";
        var config = new ClaudeTextGenerationConfig()
        {
            MaxTokensToSample = 2048,
            Temperature = 0.8f
        };

        var response = new StringBuilder();
        await foreach (var chunk in BedrockRuntime.InvokeClaudeWithResponseStreamAsync(new Model.ClaudeV2_1(), prompt))
        {
            response.Append(chunk.Completion);
        }
    
        return response.ToString();
    }

Invoke Llama 2 70B Chat

public async Task<string> GetLlmDescription()
{
    const string prompt = @"Describe in one sentence what it a large language model";
    var config = new Llama2TextGenerationConfig()
    {
        MaxGenLen = 2048,
        Temperature = 0.8f
    };
    
    var response = await BedrockRuntime.InvokeLlama2Async(new Model.Llama270BChatV1(), prompt);
    
    return response?.Generation ?? "";
}

Learn More

• Learn more about Amazon Bedrock
• Learn more about foundation model inference parameters
• C# code examples for Amazon Bedrock using the AWS SDK for .NET